Model ML is helping financial firms rebuild with AI from the ground up

Chaz Englander runs Model ML, a startup that's betting the entire financial services software stack needs to be torn up and rebuilt around AI agents rather than patched with a chatbot bolted onto the side. That's the core argument in OpenAI's latest Executive Function interview, and it's a more radical position than most enterprise software vendors are willing to state out loud.

The pitch goes like this: banks and asset managers have decades of legacy systems, spreadsheets, PDFs, and manual review processes that were designed around humans clicking through screens one step at a time. Englander's contention is that autonomous agents don't work well when you just drop them into that environment. They need infrastructure built for them from the start, meaning systems that can hand off context, verify outputs, and chain together multiple steps of reasoning without a person babysitting every stage.

Model ML's approach, as described in the piece, is to focus on financial workflows specifically, things like document analysis, research synthesis, and due diligence tasks that eat up analyst hours, and rebuild the tooling around agentic execution rather than retrofitting old dashboards with an AI feature. Englander frames this as a shift from software that assists a person to software that acts on a person's behalf, with humans reviewing outcomes rather than performing every intermediate step.

What's notable is the framing choice. OpenAI's Executive Function series is clearly aimed at showcasing customers who've gone all-in on agentic workflows, and Model ML is presented as a case study in why financial firms in particular are ripe for this kind of overhaul. Given how document-heavy and compliance-bound finance already is, it's an industry with a lot to gain if agents can reliably handle the grunt work, and a lot to lose if they get it wrong.

The interview doesn't dwell much on failure modes or how errors get caught before they reach a client, which is probably the more interesting question for anyone actually running money through these systems. Rebuilding infrastructure around agents is one thing; proving those agents are trustworthy enough for regulators and risk committees is the harder sell that this piece mostly skips past.